Deepak V.

Deepak Vaduganambi

Backend Engineer · AI / LLM Systems

I build and harden Python services, AI platform components and the seams between services, databases, workers and model providers. Focused on performance, observability, authorization and production reliability.

Get in touch GitHub

Selected impact

229×–965×

Faster permission resolution

Reworked Casbin lookups into an indexed path with byte-identical output; shipped to 17 services.

−40%

Worker boot memory

709 MB → 427 MB by removing eager heavyweight imports in a Celery enrichment worker.

15m → 35s

Enrichment step latency

Found model output-headroom limits and rerouted to a 128k-output model; rows 9/10 → 10/10.

Experience

Backend Engineer (AI Developer) · Oraczen

  • Owned and contributed to shared backend infrastructure: observability, authorization/RBAC, LLM routing, configuration, feedback and data-enrichment services.
  • Diagnosed a platform-gateway crash loop as CPU throttling and event-loop blocking (not memory), tracing probe failures to a 150m CPU limit and exit code 137.
  • Migrated shared observability to Langfuse v3 and OpenTelemetry, centralising LLM-call telemetry behind a facade enforced via CI/lint.
  • Drove per-service database separation with independent Alembic migration chains, including a two-database design for a vision service.
  • Fixed security and reliability issues: budget-enforcement bypasses, JWT forwarding, sensitive cache logging, static bearer-token fallbacks and cross-tenant authorization.
  • Built measurement harnesses and guards for memory, dependency drift and service probing rather than relying on one-off fixes.

Intern, React JS Developer · Thought Process Systems LLP

Built React / Redux Toolkit components and integrated them with backend services.

Freelance Web Developer · Accurate Info Solution

Delivered a custom WordPress site with UX improvements and SMTP email integration.

Systems I've worked on

Observability platform

Langfuse v3 migration, OpenTelemetry pipeline, Sentry/GlitchTip, analytics dashboards.

Authorization / RBAC

Casbin permission-resolution rewrite, policy reconciliation, org-vs-agent domain scoping.

LLM router

Budget enforcement, fallback bounding, reasoning-model timeouts, per-org routing.

Data enrichment pipeline

Worker memory work, run lifecycle, per-stage tracing, cost attribution at 100k+ rows/run.

Multi-tenant config & DBs

Service-owned databases with independent migration chains.

Packaging modernisation

~11 services moved to isolated Poetry environments and a standard image layout.

Skills

Backend
Python 3.12, FastAPI, Pydantic v2, Celery, Alembic, Poetry, pytest
Data
PostgreSQL, ClickHouse, Redis, RabbitMQ
Auth
Keycloak / OIDC, Casbin, JWT
AI / LLM
Langfuse, LangChain, MCP, OpenTelemetry, vector search, multi-provider routing
Infra & Ops
Docker, Kubernetes / EKS, Helm, ArgoCD, GitHub Actions, Grafana / Loki
Frontend
TypeScript, React, Next.js (supporting)

Education & certifications

Contact

Bengaluru, India · deepakrama321@gmail.com · GitHub